Extropy Security Bytes: w23 2025

Welcome to our weekly security briefing, where we delve into the notable events shaping the cryptocurrency and blockchain landscape. For week 23, we’ll be examining several incidents that underscore the dynamic nature of security in Web3, from sophisticated technical exploits to evolving social engineering tactics.

Force Bridge: Cross-Chain Bridge Experiences $3.76 Million Breach

Force Bridge, the cross-chain interoperability solution developed by the Nervos Network, suffered a significant security breach on June 1st, 2025. The attack resulted in the theft of approximately $3.76 million worth of digital assets, impacting funds held on both the Ethereum and BNB Chain deployments of the bridge.

Mechanism of Attack:

While the specific vulnerability has not been fully disclosed, analysis indicates the attacker managed to gain access control privileges. This allowed them to execute actions normally restricted to administrators or trusted parties, such as unlocking token reserves. The sources suggest this was likely achieved through compromised credentials, social engineering, or potentially an inside job, rather than exploiting a complex smart contract bug. The access granted the attacker the ability to call restricted functions like unlock(), withdraw(), or transferOwnership().

Attacker Actions and Fund Movement:

The attack appeared to be methodically prepared. The attacker funded their wallets via platforms like KuCoin the day before the exploit and made multiple failed transaction attempts before successfully executing the attack. Once access was gained, the execution was described as textbook efficient. The stolen assets included various tokens such as USDT, ETH, USDC, DAI, and WBTC. These funds were subsequently converted primarily into Ethereum (ETH). The hacker then used cryptocurrency mixers like Tornado Cash and FixedFloat to obfuscate the transaction trails, making the funds difficult to trace.

Protocol Response:

Following the detection of the suspicious activity, the Nervos Network team immediately paused the Force Bridge contracts to prevent further losses. An urgent investigation was initiated, with public communication regarding the “abnormal activity” handled by Magickbase, a partner of Nervos. Blockchain security firms such as Cyvers Alerts and Hacken flagged the incident and provided analysis.

Context and Implications:

This attack adds to a growing list of incidents targeting cross-chain bridges, highlighting them as high-value and high-risk targets within the DeFi ecosystem due to their complexity and liquidity. The scale and precision suggest a highly skilled attacker. The timing of the hack, immediately following an announcement about the protocol’s planned sunsetting, was noted as a point of interest. The incident underscores the ongoing security challenges in the decentralized finance space and the critical need for robust security measures, particularly concerning access control and key management in such vital infrastructure.

AlexLab Protocol Suffers Over $16 Million Loss on Stacks Blockchain

On June 6, 2025, Alex Protocol, a Bitcoin decentralized finance (DeFi) platform operating on the Stacks blockchain, suffered a significant exploit. While AlexLab initially reported losses of approximately $8.3 million, subsequent analysis of transaction data indicated the attacker successfully extracted over $16.18 million in various assets, including STX, aBTC, sBTC, ALEX tokens, and sUSDT. This exploit marks one of the largest in the Stacks ecosystem to date.

Mechanism of Attack:

The breach was caused by a vulnerability in Alex Protocol’s self-listing verification logic, contrary to initial claims that suggested an issue with Stacks’ transaction detection. The exploit was a systematic vault heist leveraging AlexLab’s own permission system, with the real vulnerability identified as lax vault access controls and misuse of smart contract functions.

Exploit Steps:

The attack unfolded in several methodical steps:

1. The attacker deployed a malicious token with a fake transfer function designed to drain AlexLab’s vault.
2. A legitimate pool was created, which automatically triggered AlexLab’s set-approved-token function, granting vault permissions to the attacker's malicious token contract.
3. Enabling farming on this pool unlocked the malicious token’s transfer abilities.
4. Executing a single swap call caused the vault contract to call the malicious token’s transfer function using as-contract, changing the transaction context so the vault itself appeared as the sender.
5. With vault-level permissions obtained, the fake transfer function systematically drained every asset from the vault in a single transaction to the attacker’s address. The precision and execution of the attack suggested the attacker either spent significant time studying the protocol’s code or possessed inside knowledge.

Context and Response:

This incident marked the second major security breach for Alex Protocol in just over a year, following a $4.3 million loss in May 2024 likely linked to the Lazarus Group. In response to the June 6th attack, AlexLab suspended platform activities, and the Alex Lab Foundation pledged to fully reimburse affected users using its treasury reserves, with compensation in USDC based on on-chain exchange rates.

Audit Limitations:

Notably, the protocol had undergone two security audits just three weeks prior to the exploit. While these audits identified other issues, neither caught the specific vault permission exploit leveraged in this attack, raising questions about audit scope or missed vulnerabilities. The incident highlights the ongoing security challenges in DeFi, particularly concerning access controls, smart contract logic validation, and the limitations of audits in guaranteeing complete security.

Unrestricted Large Language Models (LLMs) Threaten Crypto Security

The emergence of unrestricted Large Language Models (LLMs) presents a new and evolving threat landscape for the crypto and Web3 space. These AI systems, often fine-tuned or “jailbroken” to bypass ethical safeguards, significantly lower the barrier for launching sophisticated, large-scale, and highly deceptive attacks, enabling individuals without programming knowledge to craft malicious code, phishing emails, or orchestrate fraud.

Proliferation of Tailored Malicious Models:

Attackers can modify LLMs to target specific scams or user groups, rapidly generating highly deceptive content. Locally deployed, fine-tuned models can bypass mainstream LLM safety mechanisms.

Facilitating Phishing and Scams:

Unrestricted LLMs can effortlessly produce highly realistic and persuasive communications in virtually any language, boosting scammers’ effectiveness at scale. This enhances scam vectors like “Pig Butchering” by automating personas, crafting fluent dialogue, and scaling outreach efforts through Web3’s social infrastructure (Discord, Telegram, wallet GUIs).

Aiding Malware Distribution:

Attackers create fake project repositories on platforms like GitHub with malicious code (e.g., “GitVenom” campaign). README files for these fake projects are often well-designed, sometimes AI-generated, to build trust.

Potential for Sophisticated Exploit Development:

While not solely responsible, models trained on dark web data (like DarkBERT) could potentially be weaponized to identify crypto targets, develop exploit strategies, and automate scam operations, including writing malicious code.

Examples of Unrestricted LLMs:

Specific examples include WormGPT, marketed for business email compromise (BEC) tactics; DarkBERT, initially for security research but with weaponization potential; and Venice.ai, offering minimal safety restrictions for accelerated fraud scripting. These tools foster an underground AI ecosystem for malicious AI applications.

Countermeasures:

Countermeasures involve advancing detection capabilities for AI-generated phishing, exploits, and malware. Strengthening model defenses against misuse, developing watermarking and provenance tools for harmful outputs, and establishing ethical/regulatory safeguards are also recommended. For users, maintaining robust cybersecurity hygiene is crucial, including analyzing code, using malware protection, scrutinizing repository indicators, avoiding suspicious links, and reporting suspicious repositories. Skepticism remains a strong defense in Web3.

Crocodilus Android Trojan Targets Crypto and Banking Apps

Crocodilus, an Android banking trojan, has significantly expanded its reach and capabilities, posing a direct and growing threat to Web3 security. Initially detected in March 2025 and primarily active in Turkey, it has since launched new campaigns targeting a broader range of applications, including crypto wallets, and has expanded geographically to Europe and South America.

Mechanism and Capabilities:

The malware is spread through various means, such as Facebook Ads promoting fake loyalty apps or masquerading as browser updates. Its dropper is capable of bypassing Android 13+ restrictions. Once installed, Crocodilus employs several techniques:

• It overlays fake login pages on top of legitimate banking and crypto apps.
• It can modify infected devices’ contact lists to facilitate social engineering attacks.
• A key enhancement is its automated seed phrase collector specifically aimed at cryptocurrency wallets, allowing it to extract seed phrases and private keys with greater precision for rapid account takeovers.
• It also targets cryptocurrency mining apps and European digital banks.
• The developers have significantly strengthened its defenses through deeper obfuscation techniques to resist reverse engineering.

Threat to Web3 Security:

Crocodilus poses a direct and significant threat by directly stealing cryptocurrency wallet seed phrases and private keys, enabling fast account takeovers and asset drainage. This aligns with broader Web3 threats such as malware distribution (stealing sensitive data like private keys and exchange credentials), direct private key compromise, and sophisticated phishing/social engineering tactics. The sophistication of Crocodilus reflects the increasing technical capabilities of attackers targeting the crypto space. Its focus on directly compromising user wallets at the credential level bypasses smart contract security, highlighting the critical importance of off-chain security measures for Web3 users.

The Evolution of Web3 Phishing to AI-Driven Attacks

Phishing attacks have undergone a significant evolution, transforming from traditional online threats into a highly sophisticated danger within the Web3 space, increasingly powered by artificial intelligence (AI). Historically, phishing began with basic email spoofs, gradually incorporating personalization and evolving into “Business Email Compromise” (BEC) attacks.

Phishing in the Blockchain Era:

With blockchain, phishing adapted to irreversible payments and transparent ledgers. Early tactics included cloned ICO websites, hijacked communication channels, and malicious dApps tricking users into signing blanket approval transactions. Rug pulls became prevalent, and attackers used domain/ENS spoofing to create look-alike addresses. Phishing also targeted DAOs and governance participants with tailored spear-phishing links. A key difference is that Web3 phishing often targets off-chain generated transaction signatures rather than passwords, making rapid detection crucial due to irreversible on-chain transactions. New technical vectors include exploiting wallet protocols, social network impersonation, and blockchain naming services.

Sophisticated Web3 Phishing Techniques:

More advanced methods have emerged:

• Exchange data-breach impersonation: Leveraging extensive customer data for highly personalized campaigns (e.g., a March 2025 incident where scammers allegedly used customer data to steal over $46 million).
• Address poisoning: Sending zero-value transactions from addresses visually mimicking legitimate ones to trick users into copying the wrong address.
• Scam tokens: Unsolicited tokens containing embedded phishing URLs or triggering malicious behavior upon interaction.
• Transaction simulation spoofing: Manipulating on-chain states between wallet simulation and actual execution to make a malicious transaction appear harmless during preview, leading to full wallet drainage (e.g., a user losing 143.45 ETH in January 2025).
• Supply chain attacks: Injecting malware into legitimate software or developer tools (e.g., “GitVenom” campaign, DogWifTools, XRP Ledger developer kit compromise via phishing).
• Malware spread via fake links: Distributing malicious software through deceptive links (e.g., Zoom meeting phishing attacks).

The Rise of Generative AI:

Generative AI has dramatically enhanced these techniques. Large Language Models (LLMs) enable attackers to effortlessly generate highly realistic, persuasive, and multilingual communications for various platforms. AI automates and personalizes social engineering, making “pig butchering” scams more scalable and convincing. Deepfake videos and AI-crafted impersonations of founders or influencers can promote fake investments. Automated AI chatbots infiltrate channels, impersonating support staff. AI also enables the rapid deployment of numerous customized phishing websites.

Industry Impact:

Industry data confirms a steep rise in AI-enhanced scams post-2023, with successful incidents and losses significantly increasing. By 2025, the majority of funds entering scam addresses were reportedly connected to AI-powered schemes, with total crypto scam revenue surpassing $12 billion annually.

Illustrative Examples:

Recent incidents demonstrate these advanced tactics:

• Bybit hack (Feb 2025): Lazarus Group allegedly masked multisig signing interfaces, likely via malware.
• zkLend hacker (Feb/April 2025): The hacker who stole $9.5M subsequently lost a significant portion to a phishing scam targeting a fake Tornado Cash website.
• Eigenlayer (Oct 2024): A $6 million loss due to a compromised email thread involving an investor.
• Tapioca DAO (Oct 2024): A $4.4 million “social engineering attack” potentially linked to malware and North Korean hackers.
• Ionic Money (Feb 2025): A sophisticated social engineering attack involving a seemingly legitimate oracle and counterfeit collateral.
• UPCX hack (April 2025): $70 million stolen, likely from compromised credentials mirroring admin role exploitation.
• Lazarus Group’s modus operandi: Frequently linked to sophisticated attacks combining social engineering (fake job offers), malware distribution, and compromising private keys.

Conclusion:

This evolution shows a clear shift from exploiting simple coding errors to sophisticated attacks that leverage human vulnerabilities through increasingly convincing and automated deception powered by AI, often combined with malware or compromised infrastructure. The threat landscape is constantly adapting, making user education, vigilance, and enhanced security measures crucial for everyone in the Web3 space.

We hope this comprehensive overview offers valuable insights into the multifaceted security landscape of Web3. Staying informed about these evolving threats and continuously reinforcing security practices are crucial for the continued growth and safety of the decentralized ecosystem.

Since 2017, Extropy has been at the forefront of blockchain security, auditing smart contracts across Ethereum and Zero-Knowledge (ZK) protocols. We have collaborated with leading ecosystems, including Base, Starknet, and MINA, ensuring their smart contracts are resilient, efficient, and secure.

We specialize in DeFi, on-chain games, and ZK applications, leveraging formal verification, static analysis, and deep manual reviews to uncover vulnerabilities before they become exploits. Whether you’re working with Solidity, Rust, Cairo, or zkVMs, our collaborative approach ensures your project meets the highest security standards.

- Website: security.extropy.io

- Email: info@extropy.io

Get in touch today — let’s build safer smart contracts together!